I see, thanks for the quick clarification. So in the Method 2 there is a chance that some voxels will show up in multiple ROIs, right? Is there a modification of Method 2 that maximizes that number of labeled voxels while ensuring that they will not show up in multiple ROIs?

Thanks again, Tim

On Fri, Aug 20, 2010 at 4:15 PM, Douglas N Greve greve@nmr.mgh.harvard.eduwrote:

The difference is the partial volume correction is different if there are a bunch of other labels there (aparc+aseg) vs only one label

doug

Timothy Vickery wrote:

...

Hi all,

I'm creating binary mask volumes in a subject's native functional space from the segmented brain in aparc+aseg.mgz (FS v 4.5). I have found that doing this two different ways produces different results, and I'm wondering if anyone can illuminate why this might occur and which method is more appropriate (or what other method you would suggest)...

Method 1: Resample aparc+aseg.mgz into native functional space using mri_label2vol --seg aparc_aseg.mgz --fillthresh 0.5 [...plus the rest of the appropriate inputs such as subject's bold/register.dat]

Then I just parse the resulting image (using matlab or python code) into separate binary masks for each unique identifier that I'm using...E.g., for right IPL I load this image and create a new volume [newVol = (oldVol==2008)] and save that out.

Method 2: Create a label file from aparc+aseg.mgz for each unique identifier that I'm using, and then use mri_label2vol to produce a binary mask in native functional space: mri_cor2label --i aparc+aseg.mgz --id 2008 --l rIPL.label mri_label2vol --label rIPL.label --fillthresh 0.5 [... plus the rest of the required inputs, same as those used in Method 1]

Even though these seem like they should be equivalent to me, and although the masks produced agree for the most part, I generally get several more voxels per ROI using Method 2 than I do using Method 1 (and not complete overlap otherwise). For instance, for one subject, Method 1 yields 308 voxels in rIPL, but Method 2 yields 316 voxels; disagreement between the two occurs in a total of 26 of those voxels, so it isn't just a matter of Method 2 being more generous. The discrepancy seems to be proportional to the size of the ROI, so I get just a handful of disagreements for smaller ROIs (but it seems to happen almost all the time). Thanks for any advice on which method is better, or a suggestion of a better method.

Best, Tim

---

---

Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer

-- Douglas N. Greve, Ph.D. MGH-NMR Center greve@nmr.mgh.harvard.edu Phone Number: 617-724-2358 Fax: 617-726-7422

Bugs: surfer.nmr.mgh.harvard.edu/fswiki/BugReporting FileDrop: www.nmr.mgh.harvard.edu/facility/filedrop/index.html

Hi Doug,

OK, just one last question! I've decided I'd rather use the smaller masks (generated by resampling aparc+aseg.mgz to functional space, as in my Method 1), since the difference is small (~2%), but there's absolutely no overlap in this case. Thus it's both the most conservative approach and easiest to match up with the analyses I've already conducted using these masks outside of fs/fsfast.

The only problem is that there doesn't seem to be an obvious way to use these masks with func2roi-sess...I can produce labels from my masks, but they are in native functional space, and as I understand it func2roi-sess expects the .label file to be in the subject's 1mm^3 anatomical space (or tal). It seems I could "trick" func2roi-sess into using my masks as if they are functional contrast stat maps, but will this work? I.e., something like this:

func2roi-ses -maskanalysis DummyMaskAnalysis -maskcontrast Mask1 -maskmap Mask1.nii.gz -maskthresh 0.5 [...etc.]

Since I've set the threshold at 0.5 and the mask is binary, only my voxels should pass through. I tested this out and it actually ran with this "dummy" directory structure properly formed...Is there any obvious problem with this? (slash, obviously better way to do this?)

Thanks a lot for your help.

Best, Tim

On Fri, Aug 20, 2010 at 4:34 PM, Douglas N Greve greve@nmr.mgh.harvard.eduwrote:

you can play around with the fill threshold

doug

Timothy Vickery wrote:

...

I see, thanks for the quick clarification. So in the Method 2 there is a chance that some voxels will show up in multiple ROIs, right? Is there a modification of Method 2 that maximizes that number of labeled voxels while ensuring that they will not show up in multiple ROIs?

Thanks again, Tim

On Fri, Aug 20, 2010 at 4:15 PM, Douglas N Greve < greve@nmr.mgh.harvard.edu mailto:greve@nmr.mgh.harvard.edu> wrote:

The difference is the partial volume correction is different if there are a bunch of other labels there (aparc+aseg) vs only one label

doug

Timothy Vickery wrote:

   Hi all,

   I'm creating binary mask volumes in a subject's native
   functional space from the segmented brain in aparc+aseg.mgz
   (FS v 4.5). I have found that doing this two different ways
   produces different results, and I'm wondering if anyone can
   illuminate why this might occur and which method is more
   appropriate (or what other method you would suggest)...

   Method 1: Resample aparc+aseg.mgz into native functional space
   using
   mri_label2vol --seg aparc_aseg.mgz --fillthresh 0.5 [...plus
   the rest of the appropriate inputs such as subject's
   bold/register.dat]

   Then I just parse the resulting image (using matlab or python
   code) into separate binary masks for each unique identifier
   that I'm using...E.g., for right IPL I load this image and
   create a new volume [newVol = (oldVol==2008)] and save that out.

   Method 2: Create a label file from aparc+aseg.mgz for each
   unique identifier that I'm using, and then use mri_label2vol
   to produce a binary mask in native functional space:
   mri_cor2label --i aparc+aseg.mgz --id 2008 --l rIPL.label
   mri_label2vol --label rIPL.label --fillthresh 0.5 [... plus
   the rest of the required inputs, same as those used in Method 1]

   Even though these seem like they should be equivalent to me,
   and although the masks produced agree for the most part, I
   generally get several more voxels per ROI using Method 2 than
   I do using Method 1 (and not complete overlap otherwise). For
   instance, for one subject, Method 1 yields 308 voxels in rIPL,
   but Method 2 yields 316 voxels; disagreement between the two
   occurs in a total of 26 of those voxels, so it isn't just a
   matter of Method 2 being more generous. The discrepancy seems
   to be proportional to the size of the ROI, so I get just a
   handful of disagreements for smaller ROIs (but it seems to
   happen almost all the time).
   Thanks for any advice on which method is better, or a
   suggestion of a better method.

   Best,
   Tim

---

   _______________________________________________
   Freesurfer mailing list
   Freesurfer@nmr.mgh.harvard.edu
   <mailto:Freesurfer@nmr.mgh.harvard.edu>

   https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer

-- Douglas N. Greve, Ph.D. MGH-NMR Center greve@nmr.mgh.harvard.edu mailto:greve@nmr.mgh.harvard.edu

Phone Number: 617-724-2358 Fax: 617-726-7422

Bugs: surfer.nmr.mgh.harvard.edu/fswiki/BugReporting http://surfer.nmr.mgh.harvard.edu/fswiki/BugReporting

FileDrop: www.nmr.mgh.harvard.edu/facility/filedrop/index.html http://www.nmr.mgh.harvard.edu/facility/filedrop/index.html

-- Douglas N. Greve, Ph.D. MGH-NMR Center greve@nmr.mgh.harvard.edu Phone Number: 617-724-2358 Fax: 617-726-7422

Bugs: surfer.nmr.mgh.harvard.edu/fswiki/BugReporting FileDrop: www.nmr.mgh.harvard.edu/facility/filedrop/index.html